Mars rovers of the future will be fitted with increasingly complex context and sample analysis instrumentation. In the case of Mars 2020, the contact science payload is optimised for pet- rography. Two complimentary lithochemistry instruments on the rover’s arm are designed to generate up to tens of thousands of spectral measurements per experiment. These data are generated in addition to microscale context imaging, housekeeping data, and contextual information produced by additional sensors onboard the rover. One consequence of these technological advances is a higher fidelity experience for geologists who may be able to place themselves within virtual field environments. However, mission scientists are grappling with the same challenges as cutting edge of terrestrial investigations: the ability to generate vast datasets, from the nano- to macroscale, is advancing faster than the ability to organise, analyse and share these data. A single sample may yield gigapixel contextual images, microscale topographic models, and huge hyperspectral datasets generated by diverse analytical techniques that are difficult to cross-reference and synthesize.
For Mars rover missions, this problem is further compounded by the often-unusual characteris- tics of flight instruments, which are designed to maximise measurement flexibility and survival in harsh environments, in contrast to the ideal geometries and sample formats prevailing in laboratory set-ups. Furthermore, tight turnarounds are required for the analysis of downlinked datasets, as the time windows during which a platform can be commanded to investigate any given target are limited, and experimental results may affect key operational decisions. This is particularly true for the Mars 2020 mission, which aims to significantly improve the efficiency of rover operations in order to complete an ambitious sample selection, documentation and caching plan during the prime mission in Jezero Crater.
To address this problem, we explored novel approaches to increasing the efficiency of data analysis tasks performed by the globally-distributed Planetary Instrument for X-ray Lithochemistry (PIXL) Science Team. PIXL is an X-ray fluorescence instrument flying aboard the robotic arm of the Mars 2020 rover. In contrast to previously flown XRF instruments (e.g. XRFS, APXS) that have performed elemental chemistry measurements over regions several centimetres in diameter, PIXL will be the first such instrument to generate micron-scale elemental chemistry maps. As well as the peaks expected from the fluorescence of major and minor elements in rocks and soils, and scattered primary radiation, x-ray spectra may also include peaks related to the presence of exotic/unpredicted elements, x-ray diffraction (due to the passage of x-rays through the atomic lattices of minerals), and other spectral artifacts. PIXL scientists desire visualization tools that allow the rapid identification of features such as these that may be hidden within the very large datasets generated by the instrument.
An ongoing collaboration between astrobiologists, computer scientists, and computer-human- interaction researchers at the Queensland University of Technology (Australia) and the NASA Jet Propulsion Laboratory has culminated in a software platform (PIXLISE) that will be used to analyse PIXL data returned during Mars 2020 surface operations. PIXLISE users manipulate large, multi-spectral datasets via web browsers and a cloud-based, shareable user session architecture. Innovative data visualization approaches minimise human-in-the-loop process- ing tasks. For example, a dimensionality reduction technique known as t-SNE is used to autonomously group spectral populations into rock components, such as grains, veins and cements.
PIXLISE interfaces with PIQUANT, a quantitative XRF measurement model based on the fun- damental parameters approach developed at the University of Washington. PIXL scientists rapidly iterate quantitative calculations by passing spectral populations and model param- eters between the PIXLISE and PIQUANT instances running in the cloud. By improving the efficiency of data analysis tasks, and distributing PIQUANT computation across multiple virtual-machines, processing times for complex PIXL experiments have been reduced from days to minutes.